Forecast User Guide

“Behind good forecast practices are often hidden good theories; equally, good theories should provide a basis for good forecast practices.”     Professor Tor Bergeron, personal communication, 1974.                                 

This edition of the Forecaster User Guide applies to the ECMWF Integrated System (IFS) and meteorological products after May 2026 using IFS Cycles 50r1 and later.  Increasing guidance on the AIFS (the ECMWF data-driven/machine learning model) is also included.

Aim of the Forecaster User Guide

The aim of this User Guide is to help meteorologists make the best use of the forecast products from ECMWF.

In particular the aims of the guide are:

• to give an outline of the structure and use of the ECMWF Integrated Forecast System (IFS) and the ECMWF Artificial Intelligence Forecast Model (AIFS)

• to increase understanding of the ensemble forecast process.

• to show how the various IFS models inter-depend and interact.

• to advise on how best to use the output and how to build up trust in the forecast information.  A good forecast that is not trusted is a worthless forecast.

• to introduce and develop new products.

• to reach new sectors of society.

• to satisfy new demands.

The goal of ECMWF is to produce:

• medium-range (day 0 to day 15) forecast products.  These concentrate on detail and uncertainty probabilities of forecast events.  Presentation of this output generally differs significantly from that dealing with seasonal NWP.   

• sub-seasonal range (day 16 to day 42) forecasts products.  These concentrate on the probabilities of anomalies from the norm during 7 day forecast periods for given location and time of year.

• seasonal forecasts (month 1 to month 7 or month 13).  These give an indication of likely conditions beyond six weeks ahead.  They are run monthly giving forecasts to 7 months ahead, and run quarterly with forecasts extended to 13 months ahead.  Output concentrates on the anomalies relative to the seasonal climate.  

The IFS configurations are:

• the 15-day ensemble forecast (ENS) including the ensemble control forecast.

• the 46-day sub-seasonal range forecast including the sub-seasonal range control member.

• the 7 or 13 month seasonal forecast.

The ECMWF model output is delivered in the form of charts or GRIB format datasets.  It is readily available to forecasters via:

• imports into their own work station environments.

• on the web as Opencharts.

• using the highly interactive ecCharts web-based application (for ECMWF members and Co-operating States).

The ECMWF IFS is upgraded at roughly yearly intervals and each is identified as named cycles (e.g.Cy50r1).   These incorporate improved representation of physical processes and/or resolution changes.   New products increasingly aid early warning of severe or hazardous weather.  Information on the latest upgrade  is given below.

Structure of this guide

The User Guide is broadly divided into two parts.  Sections 2 to 5 describe the structure of the ECMWF Integrated Forecasting System (IFS).  Sections 6 to 11 describe how the IFS may be used to best advantage by forecasters.

There are links to more detailed descriptions of processes, mainly at the end of each section.  Separate online ECMWF training resources explain aspects of the ECMWF IFS more visually.

A key component of the work at ECMWF is education and training.  Further educational material is available through the web site:

• Webinars (recordings).

• Slidecasts (slides and audio recordings).

• Tutorials.

• Training lectures (presentations in PDF)). 

ECMWF Newsletters issued quarterly give information on IFS models and applications and ECMWF plans.

A glossary is included in Appendix 12C.

Section2: The ECMWF Forecasting Systems

Section 2 describes in broad, non-technical terms the Forecast systems developed and currently in use by ECMWF.

Section 2A describes the use of the ECMWF Integrated Forecast System (IFS).  This comprises the global atmospheric model, the wave and the oceanic dynamical models, and the data assimilation systems.  It gives an overview of the way the atmospheric model uses sub-gridscale parameterisations and atmospheric physics for processes within the atmosphere and at the surface.  There are large differences in energy fluxes between land or sea and the atmosphere.  Thus the definition of the model coastline by the land-sea mask is extremely important. This is especially true for meteograms in coastal areas or on islands.

Section 2B describes the use of Artificial Intelligence forecasting and the Machine Learning process where the algorithms are derived.  It gives an overview of the way forecasts are derived using observations and a well trained empirical forecast model.      Weather forecasting using Artificial Intelligence is radically different to the physics based techniques used by the Integrated Forecast System and currently widely used by forecast models elsewhere.  

Numerical weather prediction (NWP) output is complicated by its often counter-intuitive and non-linear behaviour.   Understanding model processes enables forecasters to assess model output critically.

Section3: Availability and interpolation of NWP output

Section 3 gives an overview of the way ECMWF graphical forecast products are presented to the forecaster.  It gives some insights into ways the analysed and forecast data may be reduced in accuracy by the way it is presented.

Section4: NWP evolution versus reality

Section 4 discusses model error growth with time and the relationship between predictability and scale.  An indication is given of how anomalies propagate downstream and gives some pointers towards recognition of these in the analysis.

Section5: Forecast ensemble (ENS) - rationale and construction

Section 5 describes the way the members of the ensemble are generated.  The use of ENS allows assessment of uncertainty in the model forecast by giving a range of results.  Each ensemble member starts from slightly perturbed initial data.  Consequently each evolves a little differently from the other members of the ensemble to give a range of possible forecast results.  The variation seen within the ensemble forecasts gives an indication of predictability of the atmosphere.

Model climates are an important product produced within the IFS.  These are: M-climate for medium range ensemble, SUBS-M-Climate for sub-seasonal range ensemble, S-M-climate for seasonal forecasting ensemble.  They are a wholly model-based assessment of worldwide climatology based on analyses and re-forecasts over a previous period of 20 or 30 years.  

Section6: Using ensemble forecasts

Section 6 discusses the reliance that can be placed upon the ensemble as the forecast lead-time increases.  Each slightly perturbed ensemble member evolves a little differently from the others and gives a range of possible forecast results.  The variation seen within the ensemble forecasts gives an indication of predictability of the atmosphere.  The use of probabilities or other risk assessments is an essential part of building forecasts useful to the customer.  This section emphasizes the benefit of using ensemble products to get the best description of evolution and uncertainty of a forecast.

Section7: Dealing with uncertainty

Section 7 concentrates on methods that may be used to assess confidence in model results.   This section gives guidance on interpretation of latest and previous ensembles output to allow insight into the uncertainty of the forecast.  It also gives guidance on assessing the skill of a forecast and how to use run-to-run variability in the forecasts to best advantage.  The continuing role of the human forecaster is emphasized.

Section8: ENS products - what they are and how to use them

Section 8 concentrates on making best use of the extensive range of products that are available.  The IFS produces a very wide range of products which is delivered in the form of charts or GRIB format datasets.  It is readily available to forecasters via:

• imports into their own workstation environments,

• on the web as Opencharts,

• using the highly interactive ecCharts web-based application (for ECMWF members and Co-operating States).

Model products may be deterministic or probabilistic, or in the form of anomalies from normal as defined by model climates.  Ensemble output is shown in an easy-to-use form as:

• charts, plumes, meteograms (and wave meteograms).

• charts showing the evolutions of tropical cyclones and extra-tropical depressions.

• charts giving an indication of the variability and uncertainty among the basic model forecasts.  These also compare the latest model output with its predecessors.

The model climates are used extensively to highlight locally extreme weather conditions for time of year and for forecast lead time.  The Extreme Forecast Index (EFI), pioneered at ECMWF, compares the forecast probability distribution with the corresponding model climate distribution.  The Shift of Tails (SOT) index complements the Extreme Forecast Index (EFI) by giving information about how extreme an event might be.  This is done by comparing the tail of the ensemble distribution with the tail of the M-climate or SUBS-M-climate as appropriate.

The overall aim is to allow assessment of uncertainty to provide the customer with the best and most useful guidance possible. 

Section9: Physical considerations when interpreting model output

Section 9 gives pointers towards features which can have an impact on model output.  This allows users to modify and improve forecasts for issue to customers.  Some other short-comings of the models are noted.  These will be addressed in the future but meanwhile they need to be considered by the forecaster.  It is through forecaster user feedback that important points will be identified and addressed.  The importance of critical assessment of model output by human forecasters cannot be understated.

Section10: Interfaces for displaying model output

Section 10 gives an outline of the way forecast data may be presented to the user.  ECMWF Web Charts (Open Access) give easy access to ECMWF IFS output.  The more flexible and interactive ecCharts allows users to pick-and-mix the IFS data.

Section11: Conclusion

Section 11 highlights the continuing importance of the forecaster in providing a consistent and useful product to the customer.

Section12: Appendices

Section 12 contains additional detail on statistical concepts for verifying model forecasts, the current structure of IFS, and a list of acronyms.

Comments on application of IFS and the Forecaster User Guide

The forecaster is not a computer.   Instead, the forecaster is employed to add value to model forecasts, and to identify and quantify uncertainties.  Forecasters should provide a balanced assessment of the probability of an event that is relevant to customer requirements.  

Daily operational forecasting work is largely a matter of assessing, interpreting, combining and correcting NWP information.   Also vital is the ability to identify quickly those products that are particularly relevant for a given synoptic situation.   In the medium-range especially, the use of statistical know-how counts as much as synoptic experience. 

Forecasters, and other users, should not simply follow NWP guidance.  They should act quite differently by:

• assessing the quality of data sources.  ECMWF considers the reliability of the information that it uses and rejects data that contradicts background fields.  This can be counter-productive, e.g. rejection of buoy data in the vicinity of a hurricane.  

• surveying and questioning results from many sources.

• producing forecasts with fewer details.

• assessing the uncertainty.  All forecasts have uncertainty, and that uncertainty increases with forecast lead-time.

• ideally, not giving sudden “U-turns” in guidance.   

Upgrades in latest cycle of Integrated Forecast System (IFS) model Cy50r1

Changes to product names

With effect from the introduction of Cy50r1 in May 2026, the term “HRES” is now obsolete and is replaced by the Ensemble Control Forecast.

The Ensemble Control Forecast is identical to Ensemble Control Forecast (ex-HRES) output in Cycle49 and HRES output in earlier cycles.  They have the same resolution and are scientifically, structurally and computationally identical.  Ensemble Control Forecast output, Ensemble Control Forecast (ex-HRES) output and HRES output are fully equivalent where shown in diagrams.  At the time of some older diagrams, HRES had resolution of 9km and ensemble members had a resolution of 18km.

The extended range forecast is now known as the sub-seasonal forecast.  The long range forecast is now known as the seasonal forecast.

Major model changes

Assimilation:

• New ocean and sea ice ensemble analysis system.

• Many more T2m observations than in previous cycles.

• Coupled ocean-atmosphere assimilation of microwave imagers and geostationary infrared data giving increments to ocean and sea-ice as well as upper air.

• Scale-selective EDA re-centring to address issues with tropical cyclone initialisation.

• Introduction of stratospheric humidity assimilation from radiosondes up to 60hPa.

• Upgrade of Radiative Transfer for TOVS (RTTOV) model improving simulation of satellite radiances for assimilation.

• Increased number of wave observations.

• More satellite data with higher density in time and space.  Window channels - frequency bands where the atmosphere is relatively transparent - have been included in the assimilation.

Observations:

• Atmospheric observations

  • Shorter time slots allowing more accurate comparison between model and observations.  

  • Wind tracing with ozone-sensitive data.

  • Include specific humidity increments above the tropopause.

• Ocean observations

  • In situ temperature/salinity profiles.

IFS Model:

• New ocean and sea ice model based on NEMO-SI³, with updates including:

  • Improved numerical schemes and physical parametrisation.

  • Introduction of multi-category sea ice model with forecast salinity and melt pond dynamics.

  • The more advanced sea ice model SI³ replaces LIM2.

  • Albedo diagnosed by sea ice model SI³ replaces climatological sea ice albedo.

  • Turbulence scheme for better vertical mixing processes.

  • Extension of ORCA grid area further towards the South Pole. 

• New ensemble analysis system for ocean and sea ice based on ORAS6 reanalysis.  This is used for initialising both the forecasts and the re-forecasts. The updates include:

  • More accurate representation of short-term variability (e.g. diurnal cycle of sea surface temperature).

• Wave model:  

  • New waves and sea ice interaction. Waves penetrate through open ice areas.

  • Revised wave model bathymetry.

  • Surface currents introduced into wave model with less smoothing and choppier seas.

  •  

• New glacier parametrisation scheme in ecLand.  Uses fractional ice coverage in grid square and a four-layer land-ice scheme.

• New snow parametrisation scheme for sea ice.  Uses fractional ice coverage in grid square and a single-layer snow scheme.  Reduces the warm bias seen in winter over the ice surface, especially in cloud free situations.

• Modified Stochastically Perturbed Parametrisation (SPP) configuration to reduce 10metre wind spread in the ensemble, and also related misalignment of wind speed and wave height in perturbed members.

• Model physics:

  • Convection and microphysics changes to improve the propagation of precipitation from ocean across land.

  • Updates in ordering of aerosol climatology, convection, and physics to improve tropical upper-air forecasts.

  • Reduced vertical diffusion in stable conditions in the stratosphere.

• Coupling:

  • Coupling of snow depth and sea ice thickness from the sea ice model to the atmosphere.  Allows snow cover on sea ice and variable ice thickness to be represented in the atmospheric forecast model.

  • Fully coupled ocean-atmosphere forecasts.  The model atmosphere uses sea surface temperatures (SSTs) directly from the ocean model.

Artificial Intelligence Model:

Introduction, among other features, of wave and snow cover forecasts.

More accurate representation of how sea ice affects the power of waves and the interactions between waves and ocean currents. 

 Eleven wave-related variables are being made available to users.

 AIFS v2 has learnt to perform better than earlier versions of AIFS when initialised from Cy50r1. 

Important unchanged features:

• Spacial resolution in all IFS models remains the same as in Cy49r1.

• Ensemble control forecast is disseminated using the same schedule as used by Ensemble Control Forecast (ex-HRES) in Cy49r1.

Meteorological impact

Medium range:

• Improvements to forecasts of light precipitation.

• Better representation of convective precipitation improving inland propagation of rainfall.

• Improved tropical upper-air temperature and wind forecasts (at 850hPa and 250hPa).

• Improvement of temperature and humidity forecasts around the tropopause.

• Reduction of the known SST warm bias in the Southern Ocean.  

• Improved Western Boundary Currents (e.g. Gulf Stream), and related large SST biases. 

• Dynamic evolution of marine variables (e.g. SST/SIC) in the analysis consistent with validity time.

• Forecasts of total cloud cover, dew point temperature, and 10-metre winds over sea.

• Reduced and more realistic ensemble spread of 10m wind, particularly reducing excessive near-surface wind extremes. 

Sub-seasonal range:

• Increased skill for quasi-biennial oscillation (QBO) in tropical stratospheric winds.  Stronger amplitude and more realistic vertical descent.
• More realistic stratospheric dynamics.

The sub-seasonal range ensemble forecasts (formerly extended range forecasts Cy48 and earlier) are not just an extension of the medium range forecasts but are completely separate forecast systems.  However, both start from very similar analyses.  But there are two sets of re-forecasts, one for the medium range and one for the sub-seasonal range. 

Full details of the current Integrated Forecast System (IFS) is given in the official ECMWF IFS documentation of CY50r1 (when available)

Users are advised to keep themselves updated about changes and improvements to products and model processes through the ECMWF Newsletter and web site (e.g. via the Forecast User portal)

(FUG associated with Cy50r1)

This User Guide has been compiled  by Bob Owens, with assistance from Tim Hewson, and with contributions from many other scientists and ex-forecasters at ECMWF. It is an updated version of the "User Guide to ECMWF Forecast Products" written originally by Anders Persson and published in 2011 (that had minor adjustments in 2013 and 2015).

The User Guide should be cited as follows:           Owens, R G, Hewson, T D (2018). ECMWF Forecast User Guide. Reading: ECMWF. doi: 10.21957/m1cs7h